Swann/Hik and Dahua 3MP comparison

[MaxIcon 0]

Here's the repost of the first set of Dahua/Hik comparisons from cam-it:

 

Q was kind (and trusting) enough to lend me his Hik-flashed Swann 1080p. I'll be retiring to Mexico as soon as I sell it on ebay... Anyway, I've done a few tests to see what the deal is with the 3MP setting enabled by the Hik software.

 

While it's almost impossible to tell what's going on inside the cam, we can get some real-world results by comparing the resolution of the various settings.

 

The result: The Hik 3MP setting isn't really 3MP. In the earlier thread, I calculated the scaling and figured it to be about 2.2 MP, scaled up by software to 3MP resolution, and I think that's still about accurate, but I'll have to think on that some more.

 

I'm also doing a side-by-side Dahua 2MP comparison, which has been very interesting, but it'll be a few days before I have much on that. So, on to the 3MP info.

 

First off, the Swann lens spec on the Costco website is incorrect. Whether they think it's a 4mm lens and their lens vendor is messing with them, or someone goofed up in the specs department, is hard to say, but it's clearly not 4mm, and appears to be about 3.3mm. More on that after the 3MP data.

 

The cameras are the Swann 1080p flashed with Hik firmware, and the Dahua IPC-HFW3300C set at 1080p. All internal settings are the same - 10 fps, 10 i-frames, 4096 kb/s, constant bit rate, all image enhancements (WDR, etc) off, all settings at default.

 

Here's the test setup. This is an ISO 12233 chart printed on a 6x3' banner, 6' away from the cams. Unfortunately, the printer was supposed to print it 4x3, scaled correctly with white space on one side for extra info, but they stretched it to fill the entire banner, and I didn't catch it until I had it for a while. So, it's not accurate for resolution testing, but is good for resolution comparisons. The frame is PVC, which I built many a structure from, large and small, back in my Burning Man years.

 

Hik is at the top, Dahua is at the bottom. It's very windy today, so it's held down by spare propane bottles. Both are set up for nearly identical field of view.

 

Here are the 3MP images, which show the problem - the two cameras clearly have a different field of view, even though they were matched at 1080p.

 

First step was to compare 1080p between the Dahua and Hik. Using BI, I zoomed in to the max on different parts of the chart that showed the edges of resolution, matched it between the cams, and took screenshots. You can see that the Dahua and the Hik have pretty similar resolution at 1080p, with the Dahua having a slight edge in places and the Hik in others. I believe the Dahua has a slightly better lens, but processing also affects the images - more on the lens in another post.

 

This is 3 different areas compared - again, Hik top, Dahua bottom:

 

 

 

Next, I switched both to 3MP. As we saw above, the field of view is now different, which was the first clue that something was wrong with the Hik 3MP setting. To compare, I zoomed the Dahua all the way in, then zoomed the Hik so the view matched the Dahua.

 

You can see that the Hik now has different resolution than it had at 1080p, and than the Dahua. If it was switching to a true sensor 3MP resolution, the results would be the same as the Dahua and as the previous 1080p tests, since the difference between 3MP and 1080p isn't pixels per area, but more total pixels making the image larger.

 

Same 3 areas, different zoom due to different BI view:

 

 

 

 

Finally, here are a couple of the shots side by side. The Dahua's resolution is the same on both, as it should be, regardless of the BI zoom factor, while the Hik has different resolution in 3MP than at 1080p. Presumably the Hik resolution change is due to the scaling algorithms, but again, it's hard to know what's happening inside.

 

Anyway, for those still with me, back to the Hik lens. I measure the field of view at 10' out as 18' wide, which calculates out to about 90 degrees. Traditional lens calculators say this should be a 2.7mm lens view, but they're mostly set up for 4:3 sensors, not 1080p, so if you scale it for that, you get 3.6mm. It just happens to almost match the Dahua at the minimum lens focal length setting, which is 3.3mm, so it appears to me that the Hik lens is really around 3.3mm, maybe a bit less. I haven't looked to see if there are any markings yet...

[Swann Tech 0]

Hi guys,

 

I can confirm for you that the current Swann IP camera available from Costco is not a 3MP sensor.

 

I don't have any details about the lens but if MaxIcon's calculations are correct it could possibly be a 3.2mm lens. Even so the edge curvature of the image doesn't adequately suggest 3.2mm but I haven't had the opportunity to compare cameras in this resolution before.

[zikronix 0]

Hi guys,

 

I can confirm for you that the current Swann IP camera available from Costco is not a 3MP sensor.

 

I don't have any details about the lens but if MaxIcon's calculations are correct it could possibly be a 3.2mm lens. Even so the edge curvature of the image doesn't adequately suggest 3.2mm but I haven't had the opportunity to compare cameras in this resolution before.

 

 

Great to know...thanks for the info. Im still getting around to moving those cameras around on that soft right side. I printed the star pattern to help

[MaxIcon 0]

OK, here's my analysis on the Hik/Swann 3MP sensor resolution. Thanks to Mark at cam-it.org for posting a great high-res pic of the sensor that really helped with this, and for Q's and Buellwinkles pics of the 1080p and 3MP images.

 

tl;dr version:

 

The Hik sensor appears to be about 2.5MP, with sensor resolution of 1275x1920, for an aspect ratio of 3:2. Assuming the 1080p image is true 1080p, the 3MP image is actually a 2.2MP image, scaled up 120% to give an apparent 3.2MP image (note that MP scales with the square of the horizontal and vertical resolution).

 

I could be wrong, but all the data works together pretty well, if you want to slog through the rest of the post. If anyone sees any flaws in the reasoning, I'm always open to corrections.

 

 

First, a couple of assumptions, because you have to start somewhere:

 

- The Hik and the Swann are mechanically and electronically identical, with just minor firmware/NVRAM differences. Based on Buellwinkle's images from a Hik, this is a pretty sure thing.

 

- The Hik is a true 1080p camera - that is, when it puts out 1920x1080 pixels, they came from 1920x1080 sensor elements. We'll assume Swann's marketing and specs are true and correct, and that Hik's 3MP specs are in question.

 

- The sensor size is 1/2.5". One complication in all this is the Hik sensor size spec. The web page says it's 1/2.5" (0.4") diagonal, while virtually all the online docs (PDFs, flyers, etc) say 1/3" (0.33") diagonal. The data shows it's likely to be 1/2.5", and it looks like they updated the web page but not all the rest of the data.

 

ETA: This was a mis-assumption, as I forgot that sensor size listed isn't actually the real sensor size, but is about 50% larger than actual sensor size, due to the history of sensor construction. This doesn't affect the resolution and aspect ratio results, but means the physical size is different than calculated.

 

Here are the image specs from Hik:

 

1080p = 1920x1080 (2.1 MP - this is 1080p by definition)

3MP = 2048x1536 (3.2 MP - there are no 3MP standards, so this is what they chose)

 

Most of the sample images below are reduced to 1/3 their original size.

 

 

For starters, here are Q's original images that started me wondering about this. Since the 3MP image specs show more pixels in both axes, the 3MP should have more horizontal image showing, but it actually has less.

 

This says that the 3MP 2048 horizontal resolution is really less than the 1080p's 1920 pixels. The 3MP vertical 1536 resolution is larger than the 1080 resolution, so it is seeing more than 1080 pixels vertically.

 

Not only that, switching between both images shows the 3MP is zoomed larger than the 1080p, and if it was capturing actual sensor elements, everything common between the two images should be the exact same size, with more image around the edges of the 3MP image.

 

Here are these images, reduced by 1/3. The key is to see where each image stops compared to the other, as that's where the captured sensor pixels stop.

 

1080p image:

 

 

3MP image:

 

 

How to figure out what the real resolution is? Again, assuming the 1080p is correct, the easy way is to scale down the 3MP and overlay it on the 1080p until they match. This happens at about 83%, as seen below. If you scale down the horizontal and vertical pixel counts by 83%, you get this, which is what I posted in the original thread:

 

3MP actual = 1700x1275 (2.2 MP)

 

Here's the 83% reduced 3MP image overlaid on the 100% 1080p image.

 

 

 

We assume that the sensor maximum horizontal resolution is 1920, since that's the 1080p horizontal resolution, and we then assume the sensor maximum vertical resolution is the 1275 that the 3MP uses, because if they had more, they'd use it, you'd expect.

 

This gives us a sensor resolution of:

 

Hik sensor = 1920x1275 (2.5 MP) - note that this isn't 4:3 (SD) or 16:9 (HD), but is about 3:2.

 

Now, it's very difficult to tell what's going on in the software, so the next step is to look at the actual sensor and see if it fits. Cam-it member Mark posted a huge pic of the sensor mounted to the board - here's a reduced version.

 

 

 

This is a very useful picture, as we can tell several things from it:

 

- The aspect ratio of the imaging section of the sensor chip

- The relative size of the chip compared to the SMD resistors, which come in standard sizes

 

First step is to determine how many pixels wide and tall it is. This is easy in an image editor; by selecting the imaging area, you just look at the pixel size of the selection. It turns out that the aspect ratio is about 1.5:1, or 3:2 - the same size as the estimate from the image scaling above. This is good, because it helps verify the initial estimate!

 

Next, the resistors are measured. It's hard to measure them accurately end to end, because of the solder, but most smaller SMD resistors are twice as long as they are wide, and the width is easy to measure. This gives us the length of a resistor in image pixels.

 

Since we know how many image pixels the sensor is, we can then calculate how many resistors tall and wide it is, and it happens to be about 5.1 resistors tall and 3.4 resistors wide.

 

SMD resistors come in various standard mm lengths, like 0.6mm, 1.0mm, 1.6mm, etc. If we convert the number of resistors to mm length, then convert that to inches, that gives us several possible sizes for the sensor's horizontal and vertical dimensions.

 

Since we know the diagonal is either 0.33" or 0.4" (and suspect 0.4" is correct), we can calculate the diagonal from X^2 + Y^2 = Z^2. Plugging in the various resistor standard lengths, a 1.6mm resistor gives us a sensor size of 0.39", which is plenty close enough to 0.4, given the sketchy measurements.

 

ETA: Due to the physical image size being roughly 2/3 the listed image size (see a few posts below), this isn't right. It's more likely these are 1mm resistors, and it just happened the 1.6mm size fit in the sensor size scaling. The other conclusions are still valid.

 

So, using a completely independent set of measurements, this supports the original 3MP scaled estimates done by overlaying the images. At this point, my confidence in the results is pretty good.

Edited May 26, 2013 by Guest

[Q2U 0]

OK, here's my analysis on the Hik/Swann 3MP sensor resolution...

 

Wow. This is an MVP (MaxIcon Valuable Post); thanks a for all the good work. " title="Applause" />

Edited May 25, 2013 by Guest

[tome9999 0]

This is not at all surprising, especially with the cheaper cameras from asian manufacturers. Back in the earlier days of DSLRs, camera manufacturers res'd up images off of smaller sensors and marketed them (aka, lied) as higher res cameras than they really were. Not until people started analyzing them on camera review sites and calling them out on it did this stop happening. Now the reverse is true, manufacturers are often using slightly over sized sensors and downsampling or selective sampling to get advertised resolutions that are slightly lower than the sensor capability.

 

Good job on posting this info and keep it up! Perhaps the camera manufacturers (even the Chinese ones) will stop lying about their actual resolution.

[MaxIcon 0]

It is a little disappointing (and may be why Swann only markets these as 1080p), but what's puzzling is the odd sensor size. This isn't a set of specs usually used for ip cams, unless this is a common size used for some other apps.

 

The 3:2 aspect ratio is what traditional 35mm film uses, but these sensors don't usually have the properties of a video sensor, and tend to be expensive.

[tome9999 0]

The 3:2 aspect ratio is what traditional 35mm film uses, but these sensors don't usually have the properties of a video sensor, and tend to be expensive.

 

The current generation, last several years, of sensors used in dslrs and P&S cameras are fabulous at video and are not that expensive (hence price decreases and higher and higher resolutions). They have better specs for video (and still) as any of the best dedicated video sensors of years back. Current dslrs low light performance, still and video, is astounding. I would take a sensor from a Canon or Nikon dslr or even P&S for one of these video cameras any day!!! These sensors ARE slowly trickling into these cameras as we have seen with Sony Exmor sensors showing up.

[MaxIcon 0]

Yeah, this may be what's happening, though we've seen the Hik sensor isn't great at low light, but that could account for the lower price. Thinking on it more, many still cams now have 1080p/720p video capability, so that would be built in to the chip. The only fly in the ointment is the relatively low resolution - 2.5MP is very low for a still cam these days, but they may tether sensor elements together for video output to get better low light performance. Sounds like some research is in order.

 

I realized I made an error in one of my assumptions - it doesn't affect the results, but the sensor size (in mm/inches) calculation is wrong.

 

Sensor sizes are misleading, due to the history of sensors as originally embedded in glass tubes, where the size was the tube internal diameter, not the sensor internal diameter. So, a 1/3" sensor is actually smaller than 1/3" on the diagonal, generally by about 1/3, but there are no specific rules for this.

 

This is discussed in a lot of places, and here's one:

http://www.dpreview.com/news/2002/10/7/sensorsizes

 

This would mean my physical size calculation is off by roughly 1/3, but the resolution and aspect ratio calcs are the same. I knew about this, and have posted it before, but didn't consider it in this case.

[Q2U 0]

...I knew about this, and have posted it before, but didn't consider it in this case...

 

 

I'm beginning to think that perhaps Mr. MaxIcon is some kind of giant-brain benevolent alien from deep space on a mission to help Earth humankind jump its knowledge-base into The Next Level.

 

[MaxIcon 0]

Heh! No, just an aspie with many decades of coping strategies under my belt. We make the best engineers as long as we care about whatever we're doing, and understanding the technology behind camera sensors and optics is a current hot button for me. This, too, shall pass.

 

[MaxIcon 0]

I finally got around to doing a similar comparison on the Dahua, after some discussion on another forum, and rolled up the differences into a simple comparison. Here are the Hik 3MP and 1080p images overlaid so that the pixels match:

 

 

And here's the Dahua, with the same overlay:

 

 

Here's the simple, real-world rollup on what the specs say versus what the cameras give.

 

Hik 3MP going from 1080p (1920 x 1080) to 3MP (2048 x 1536):

7% increase in horizontal pixels, 20% decrease in field of view

42% increase in vertical pixels, 18% increase in vertical field of view.

 

Dahua 3MP going from 1080p (1920 x 1080) to 3MP (2048 x 1536):

7% increase in horizontal pixels, 7% increase in field of view

42% increase in vertical pixels, 42% increase in vertical field of view.

 

It's not at all clear what Hik's doing at the sensor level, but the actual output says they're doing something unusual somewhere.

 

I'd be interested in seeing similar comparisons between 3MP and 1080p on other 3MP cams, since these are the only 2 I've tested. More data is always good.

[cctv-dave 0]

This is a list of some of the sensors used by Hik, taken from their firmware for the DS-2CD8254FWD.

I need to obtain one of the units and have more of a poke around. Somewhere in there will be a TTL port and I'll wager given the content of the firwmare file that the unit will spit out what sensor is fitted during boot up, becasue most DM365/368 based products (or cams that run linux TBH) do.

 

MT9P031	5MP	1/2.5	2592x1944	aptina
AR0331	3.1MP	1/3	2048x1536	aptina
MT9D131	2MP	1/3.2	1600x1200	aptina
MT9M033	1.3MP	1/3	1280x960	aptina	AKA	AR0130
MT9V136	PAL/VGA	1/4		680H x 512V	aptina

A3372E3	2.1MP	1/3	1920x1080	AltaSens

IMX122	2.43MP	1/2.8	1984×1225	sony
IMX035	1.3MP	1/3	1329x1049	sony
EFFIOP	960H CCD DSP	sony
EFFIOE	960H CCD DSP	sony
CXD3191			sony ?

MN34041PL	2.1MP	1/3	1944x1092	Panasonic

a9m033	???
SS3		??

 

 

It's not an exhaustive list and there are a couple of unknowns in there. The newer bullets released in August dont have frimware uploads at present on Hik Euro site.

 

 

No surprises all the refrence design sensor and processor combinations are in there with the usual sensor suspects.

 

An awful lot of manufactuers are using essentially the same reference designs. The difference comes with the software and there are massive differences between skills of whomever is writing this for different companies.

I've seen good software ported to what was a bad camera and big changes in functionality, image quality and performance.

 

Some OEM's are quite good at that...

 

 

It's still a question which sensor they are using or how exactly they are doing it. They could be using a combination of row and column binning and skipping.

This could give different results if in 3MP mode or 1080 mode.

Or just as many sensors produce a 1080p output from more pixels than required and software scale the image, there could some skulduggery taking place.

 

Aptina do an essentially VGA sensor (IIRC MT9V136) which scaled up to PAL/NTSC equivilent resolution.

 

Depends how much you want to get into it to find out

[zikronix 0]

I tried to ttl from one of the swann varifocals from costco. I couldnt get any data out of it. Maybe my ttl adapter is borked. I seem to recall on that particular camera though that it was an aptina

[Brenning 0]

Reading through all this info, the burning question from me is: Have i just purchased a load of cameras (hik DS-2CD2032-I's) that actually arent 3mp despite being advertised at 3mp?

 

Brenning

[MaxIcon 0]

This depends on your definitions and on what Hik is doing at the sensor level.

 

The 3MP mode puts out a 3MP sized image, so by that definition, yes, it's 3MP. Of course, you could take a VGA image and upscale it to 3MP to fit that definition.

 

It's possible the 3MP is a true 3MP image, sampling one sensor pixel for each image pixel, and the 1080p image is the sampled one. If this is the case, the sensor is larger than 3MP, and the 1080p is sampling a larger-than-1080p image down to 1080p.

 

This is a legitimate way to do these things, though most vendors don't do this.

 

It's also possible that the 1080P is a true one-sensor-pixel = one-image-pixel image, in which case the 3MP is sampled from a smaller image.

 

I don't know any way to determine which they're doing except by digging through their code, and that's not on my top 10 list of things to do.

 

Here's what it boils down to, though:

 

For the money, this is one of the best cameras in its class right now. As long as you understand the limitations of the image modes and it still fills your needs, everything's fine. If you need a larger FOV at 3MP than at 1080P, the way most cameras with similar specs do, this won't work for you.

 

I could use the extra horizontal FOV on my 2 Hik cams, but I was aware of the limitation when I bought them, so no complaints. The other option was buying Dahua, and I've sworn off of them.

[Wallboy 0]

The Dahua seems to only support 2 streams according to both Q-See and Dahua's manual. The mainstream and the D1 max res limited substream. The Hikvision cameras manual says it supports up to a third stream. I was wondering what settings you can configure this third stream for. Is it possible to have both a 1080p and 720p stream coming from these Swann/Hik cameras?

[MaxIcon 0]

The latest firmware for the Dahua 2100N, 3300C, and Hik DS-2CD2032-I all only support two streams with D1 max on the second stream. That could change with any firmware release.

 

You could try sending a request to them for this feature. Dahua doesn't seem to pay much attention, but Hik may be better.

[Brenning 0]

Is the 3rd stream possibly in reference to the RTSP? Or are we talking 3 different streams at different resolutions etc.